Correlation between T-PSA, Gleason Score and the result of [¹⁸F]-FCH PET/CT including SUV_max in patients with prostate cancer: retrospective analysis.

Aim: The aim of this analysis was to assess the correlation between T-PSA, Gleason score and the results of [¹⁸F]-FCH PET/CT including SUV_max in prostate cancer patients using conventional statistical approach and an artificial neural network in parallel.

Methods: Retrospectively 754 [¹⁸F]-FCH PET/CT studies were analysed, where 169 were performed for staging and 585 because of suspicion of recurrence of prostate cancer. The data was evaluated using conventional statistical approaches and an artificial neural network. There were 19 input variables in 754 patients using unsupervised learning of data representation to uncover associations.

Results: The T-PSA was statistically higher for staging patients compared to recurrence patients (58.4 vs. 23.0; p=0.0002). The prostate SUV_max was statistically higher for those with pelvic lymph node metastases (LN) compared to those without (7.7 vs. 6.5; p=0.013). There was a statistically significant higher SUV_max for staging patients compared to recurrence patients (8.4 vs. 5.7; p<0.001). In the staging group, statistically significant differences were only noted for T-PSA in patients with more than 5 bone metastases (245.5 vs. 38.8; p<0.001). In contrast, the recurrence group demonstrated statistically significant increases in T-PSA for those with specific disease sites including: pelvic LN (41.3 vs. 13.3; p=0.0007), abdominal LN (64.8 vs. 14.8; p<0.001), thoracic LN (64.8 vs. 15.1; p<0.001), H&N LN (112.5 vs. 18.6; p<0.001) and more than 5 bone metastases (113.3 vs. 9.1; p<0.001). T-PSA was statistically higher for those with disseminated disease than clear (p<0.001), local (p=0.003) or LN (p=0.001) based disease with no other relationships noted. Further analysis was undertaken using an artificial neural network and a binary classification of disseminated disease (cumulative bone, disseminated and organ metastases) or local (lymph node, local, prostate and clear). The strongest correlation with the binary classification were single bone metastases (0.383), more than 5 bone metastases (0.361) and 2-5 bone metastases (0.348). T-PSA (0.254), SUV (0.0178) and Gleason score (0.114) ranked lower. The final architecture of the neural network correlated with a sensitivity of 65.4% and specificity of 53.8% and this is reflected in area under the curve of 0.538 (ROC analysis).

Conclusions: Prostate cancer is more likely to be disseminated if the patient is independently and collectively: under 60 years, Gleason score above 7, T-PSA above 60, SUV_max above 12, or in patients presenting scanned for recurrence.